Regulation of CD8+ T cells infiltration and immunotherapy by circMGA/HNRNPL complex in bladder cancer

The limited success of immunotherapies targeting immune checkpoint inhibitors is largely ascribed to the lack of infiltrating CD8+ T lymphocytes. Circular RNAs (circRNAs) are a novel type of prevalent noncoding RNA that have been implicated in tumorigenesis and progression, while their roles in modulating CD8+ T cells infiltration and immunotherapy in bladder cancer have not yet been investigated. Herein, we uncover circMGA as a tumor-suppressing circRNA triggering CD8+ T cells chemoattraction and boosting the immunotherapy efficacy. Mechanistically, circMGA functions to stabilize CCL5 mRNA by interacting with HNRNPL. In turn, HNRNPL increases the stability of circMGA, forming a feedback loop that enhances the function of circMGA/HNRNPL complex. Intriguingly, therapeutic synergy between circMGA and anti-PD-1 could significantly suppress xenograft bladder cancer growth. Taken together, the results demonstrate that circMGA/HNRNPL complex may be targetable for cancer immunotherapy and the study advances our understanding of the physiological roles of circRNAs in antitumor immunity.


INTRODUCTION
Bladder cancer is one of the most prevalent urinary system malignancies worldwide [1]. Approximately 70% of newly diagnosed cases present with the non-muscle-invasive bladder cancer (NMIBC), while 30% of patients are diagnosed as muscleinvasive bladder cancer (MIBC). Although most NMIBCs can be managed with curative intent, about 15-20% of NMIBCs progress to MIBCs after repeated recurrences [2]. Clinically, cisplatin-based chemotherapy followed by radical cystectomy remains the firstline treatment of choice for MIBCs [3]. Unfortunately, around 50% of MIBCs are ineligible for cisplatin treatment, and nearly all MIBCs will eventually progress and die from their disease despite the initial response associated with cisplatin-based regimens [1]. Thus, it is urgent to illuminate the pathogenesis of bladder cancer and seek new therapeutic approach, thereby improving survival outcomes in patients with bladder cancer.
Immunotherapies, especially programmed cell death 1(PD-1)/ PD-1 ligand (PD-L1) checkpoint blockade, have changed the landscape of cancer treatment by producing durable responses through triggering long-term antitumor immunity. However, only a minority of patients respond to the current immunotherapy treatment [4]. Mounting evidences suggest that responses to immunotherapy rely upon tumor infiltration by T lymphocytes, specifically CD8 + T cells that are able to recognize and kill cancer cells [5]. In MIBC, intratumoral CD8 + T cells are a significant positive predictor of clinical outcomes [6]. Of note, chemokines, through interaction with the corresponding receptors, play critical roles in controlling immune trafficking and tumor infiltration [7]. The optimal CD8 + T cells infiltration is mainly determined by CCL3, CCL5, CCL20, and CXCL10, which can be produced at the tumor site [7][8][9]. Therefore, elucidating the mechanisms leading to CD8 + T cells infiltration is of paramount importance to improve the responsiveness to immunotherapy in bladder cancer.
CircRNAs are predominantly the products of back-splicing events that splice an exon to a preceding exon rather than to a downstream exon, resulting in covalently circularized RNA molecules. The landscape and action of circRNAs are systematically altered in cancer [10,11], demonstrating the important influence of circRNAs on tumorigenesis and progression. In bladder cancer, we have uncovered that circRNAs affect the biological function by sponging miRNAs [12,13]. In addition, we recently identified circRNA as an interacting partner of protein to regulate cisplatin chemotherapy for bladder cancer [14,15]. Notably, circRNAs participate in the activation of anti-tumor immunity via the recruitment of immune cells in non-small cell lung cancer and head and neck squamous cell cancer [16,17]. Notwithstanding, further efforts are still demanded to address the regulatory functions as well as complexities of circRNAs in modulating immune cells infiltration and immunotherapy, and whether circRNAs could associate with antitumor immune response in bladder cancer have not yet been explored.
Heterogeneous nuclear ribonucleoprotein L (HNRNPL), first identified as a member of the hnRNP family [18], is a type of RNA-binding protein involved in multiple cellular processes, including mRNA stabilization [19,20]. Recent studies have revealed functions of HNRNPL in various diseases [21][22][23]. However, whether HNRNPL interacts with circRNAs and its role in immunotherapy of bladder cancer remain to be further investigated.
Herein, we discover that a circRNA hsa_circ_0000591, named as circMGA, is significantly downregulated in bladder cancer and positively correlated with favorable survival. Our study provide evidence that circMGA increases the expression and secretion of CCL5 by interacting with HNRNPL, augmenting the intratumoral infiltration of CD8 + T cells and response to PD-1 checkpoint blockade. Intriguingly, HNRNPL in turn increases the stability of circMGA, forming a feedback loop that enhances the function of circMGA/HNRNPL complex. Taken together, these findings unveil the clinical impact, biological roles and underlying mechanisms of circMGA and lead us to propose that circMGA/HNRNPL complex function as an effective regulator to boost the efficacy of immunotherapy in bladder cancer.

Characteristics of circMGA in bladder cancer cells
Previous RNA-seq analysis uncovered a plethora of differentially expressed candidates between normal and cancerous bladder tissues [12]. These circRNAs were then screened in enlarged number of paired tissue samples. CircMGA (hsa_circ_0000591) was downregulated in two paired bladder cancer samples, confirmed in other 73 paired samples of bladder cancer (Fig. 1A), while the expression of MGA pre-mRNA (pMGA) and mRNA (mMGA) showed no significant difference (Fig. S1A, B). Downregulation of circMGA was also confirmed in human muscle-invasive bladder cancer cells EJ, T24T and UMUC3, compared with human immortalized uroepithelium cells SV-HUC-1 (Fig. 1B). Then, the correlation between circMGA expression and clinicopathologic features in patients with bladder cancer was analyzed and we discovered that the expression of circMGA was negatively associated with bladder cancer pathological stage and grade (Table S1). Furthermore, the preliminary Kaplan-Meier analysis illustrated that patients with low levels of circMGA were more likely to be exhibited with poor overall survival (OS) (Fig. 1C), while similar OS was found between different expression levels of mMGA (Fig. S1C). Taken together, circMGA downregulation was not simply a by-product of splicing and was negatively correlated with more advanced clinical stage of bladder cancer. Therefore, these observations prompted us to investigate whether circMGA was functional.
The genomic structure indicates that circMGA is generated from the exon 2 of MGA gene with a length of 1131 nt. The back-spliced junction site of circMGA was amplified using divergent primers and confirmed by Sanger sequencing, and the sequence of PCR product was consistent with circBase database annotation ( Fig.  1D; Fig. S1D) [24]. PCR analysis revealed that circMGA was only amplified by divergent primers in cDNA, but no product was amplified in genomic DNA (gDNA) (Fig. 1E). Next, we used RNase R exonuclease to pretreat RNA. As expected, circMGA was resistant to RNase R, while MGA mRNA was significantly reduced after RNase R treatment, which indicated that circMGA harbored a circular RNA structure (Fig. 1F). Furthermore, circMGA transcripts were more stable in comparison to MGA mRNA upon treatment with actinomycin D (ActD) in bladder cancer cells and SV-HUC-1 cells ( Fig. 1G; Fig. S1E). In addition, we performed nuclear and cytoplasmic fractionation and fluorescence in situ hybridization assay (FISH) to further obtain the subcellular localization of circMGA, and the results implied that circMGA was mainly localized in the cytoplasm (Fig. 1H, I; Fig. S1F). Collectively, these data established that circMGA, derived from exon 2 of the MGA gene locus, was a bona fide circRNA, which was highly stable and located in the cytoplasm.

CircMGA increases the accumulation of CCL5
To explore the function of circMGA, we constructed circMGA overexpression plasmid, and confirmed circMGA was overexpressed efficiently in bladder cancer cells (Fig. S2A). Meanwhile, overexpression of circMGA did not alter the mRNA levels of MGA ( Fig. S2A), demonstrating the fidelity of the overexpression systems used to manipulate circMGA. To our surprise, cell proliferation, apoptosis, cell cycle progression, and cell metastasis were not obviously affected upon enforced expression of circMGA ( Fig. S2B-F).
To further elucidate the potential function and mechanisms underlying of circMGA, transcriptome analysis was performed ( Fig.  2A). Among these differentially expressed mRNAs, four mRNAs were upregulated in all three stable transfectants cells, whereas no consistently downregulated gene was observed (Fig. 2B, C). Notably, only CCL5 was increased at both mRNA and protein levels ( Fig. 2D-F). Meanwhile, we performed analysis in the TCGA database, and found that high level of CCL5 was associated with better OS (Fig. 2G). Furthermore, the down-expression level of CCL5 was noted in tumor tissues (Fig. 2H). In addition, high expression level of CCL5 was remarkably correlated with good prognosis of bladder cancer patients (Fig. 2I). Importantly, expression of circMGA was significantly positive correlated with CCL5 ( Fig. 2J). Altogether, these results demonstrated that circMGA could up-regulate the expression of CCL5 in bladder cancer cells, and that high levels of CCL5 predicted a longer survival time for OS.
CircMGA interacts with HNRNPL protein in bladder cancer cells CircRNAs have been reported to act as the sponge to miRNAs or function by interacting with proteins [25]. Given that circMGA was abundant and stable in cytoplasm of bladder cancer cells, we wondered whether circMGA regulated the target gene CCL5 by binding miRNAs. Surprisingly, circMGA was not specially enriched by AGO2, indicating that circMGA might not function as miRNA sponge to regulate the expression of CCL5 (Fig. 3A). We then performed RNA pull-down assays and mass spectrometry to define its protein binding role, using biotin-labeled probes targeting the circMGA back-spliced sequence (Fig. 3B). Following the analysis pipeline to identify RNA-binding proteins (RBPs), a major differential precipitated band was identified to be HNRNPL (Fig. 3C, D; Table S3). The interaction of circMGA with HNRNPL was further demonstrated through RNA pull-down assay and RIP analysis (Fig. 3E, F; Fig. S3A). In addition, we validated endogenously expressed circMGA colocalized with HNRNPL in the cytoplasm using immunofluorescence and fluorescence in situ hybridization assays (Fig. 3G).  1 Identification and distribution of circMGA. A The expression of circMGA in 73 pairs of bladder cancer and paired adjacent normal bladder tissues was determined by qRT-PCR. GAPDH was used as internal control. Data were mean ± SD, paired t-test. B The expression of circMGA was detected by qRT-PCR in SV-HUC-1, EJ, T24T and UMUC3 cells. GAPDH was used as internal control. Data were mean ± SD, n = 3. ***P < 0.01 versus SV-HUC-1, Student's t-test. C Overall survival (OS) of bladder cancer patients was analyzed by Kaplan-Meier curves. Patients were grouped by the median circMGA expression. P value was calculated using a log-rank test. D Schematic illustration showing the formation of circMGA. The back-splice junction site of circMGA was validated by Sanger sequencing. E The existence of circMGA was validated in three bladder cancer tissues and SV-HUC-1, EJ, T24T, and UMUC3 cell lines by qRT-PCR. GAPDH was used as a linear control. Red arrows indicated divergent primers, and blue arrows indicated convergent primers. F Analysis for RNA levels of circMGA and MGA mRNA by qRT-PCR in EJ, T24T and UMUC3 cells treated with or without RNase R. Data were mean ± SD, n = 3. ns means no statistical significance, ***P < 0.01, Student's t-test. G Analysis for RNA levels of circMGA and MGA mRNA by qRT-PCR in bladder cancer cells treated with actinomycin D (ActD) at the indicated time points. n = 3. H Identification of circMGA cytoplasmic and nuclear distribution by nuclear and cytoplasmic fractionation assay in bladder cancer cells. GAPDH and U1 were applied as positive controls in the cytoplasm and nucleus, respectively. n = 3. I Identification of circMGA cytoplasmic and nuclear distribution by RNA fluorescence in situ hybridization (FISH) in EJ cells. 18S and U6 were used as positive controls in the cytoplasm and nucleus, respectively. circMGA, 18S and U6 probes were labeled with Cy3, Nuclei were stained with DAPI. Scale bar, 10 µm.
An anti-Flag RIP assay demonstrated that the RRM3-RRM4 tandem domains, but not other domains, were crucial for its interaction with circMGA ( Fig. 3I, J). In summary, these results proposed that circMGA/HNRNPL formed an RNA-protein complex through the RRM3-RRM4 tandem domains of HNRNPL in bladder cancer cells.
HNRNPL enhances the stability of circMGA and CCL5 Given that HNRNPL is essential for RNA stability [19,22,28] and that circMGA could interact with HNRNPL, we asked whether HNRNPL increased the stability of circMGA. It showed that enforced expression of HNRNPL efficiently upregulated the expression of circMGA ( Fig. 4A; Fig, S3B). Conversely, knockdown of HNRNPL reduced the expression of circMGA ( Fig. 4B; Fig, S3C). However, overexpression of circMGA had no effect on either mRNA or protein levels of HNRNPL (Fig, S3D, E). As the target of circMGA, the expression and secretion of CCL5 were also markedly increased or suppressed by ectopic expression or knockdown of HNRNPL ( Fig. 4A-D).
It was previously reported that HNRNPL could affect RNA stability through binding to the 3′-untranslated region (3′-UTR) [28]. Alternatively, we found both circMGA and CCL5 mRNA stabilized by HNRNPL (Fig. 4E, F). As demonstrated by dual-luciferase reporter assays, a significant increment in luciferase activity of CCL5 3′-UTR reporter was detected in bladder cancer cells overexpressing HNRNPL (Fig. 4G). Additionally, CCL5 was specially enriched by HNRNPL, suggesting the interaction between CCL5 and HNRNPL ( Fig. 4H). To delineate which domain of HNRNPL contributes to the interaction with CCL5 mRNA 3′-UTR, we used catRAPID algorithm for RNA-protein interaction prediction [24]. RNA region (326-402 nucleotides) in the 3′-UTR of CCL5 mRNA was predicted to mainly bind with the 189-240 amino acids of HNRNPL (Fig. S3F). An anti-Flag RIP assay demonstrated that the RRM2 tandem domain, but not other domains, was crucial for its interaction with CCL5 mRNA 3′-UTR (Fig. S3G, H). Additionally, we constructed a fragment of CCL5 mRNA 3′-UTR with deletion of the RNA region (326-402 nucleotides, named as CCL5 Δ326-402 3′-UTR), and inserted it into downstream of the luciferase reporter gene, and validated that HNRNPL increased the luciferase activity of CCL5 mRNA 3′-UTR, but not the CCL5 Δ326-402 3′-UTR ( Fig. 4G; Fig. S3I). These results suggested that HNRNPL mainly bound with the RNA region (326-402 nucleotides) in the 3′-UTR of CCL5 mRNA through the RRM2 tandem domains in bladder cancer cells. Taken together, these findings determined that HNRNPL combined with circMGA and CCL5 to increase their expression by enhancing the stability of RNA.
CircMGA increases the expression of CCL5 through HNRNPL We returned to consider how circMGA regulated the accumulation of CCL5. Considering that HNRNPL was crucial for CCL5 mRNA stability and that circMGA could form an RNA-protein complex with HNRNPL, the influence of circMGA on CCL5 mRNA stability was examined. We confirmed that circMGA significantly stabilized CCL5 mRNA upon ActD treatment (Fig. 5A). Consistently, ectopic expression of circMGA could enhance luciferase activity of CCL5 3′-UTR reporter but not CCL5 Δ326-402 3′-UTR ( Fig. 5B; Fig. S3J). These results indicated that the function of circMGA in upregulating CCL5 expression was abolished in the absence of CCL5 mRNA-HNRNPL interaction. Importantly, overexpression of circMGA could increase the binding between HNRNPL and CCL5 (Fig. 5C). Besides, we tested the expression level of CCL5 in circMGA-KD bladder cancer cells. As shown in the results, knockdown of circMGA indeed significantly reduced the expression and secretion of CCL5 ( Fig. S4A, B). Moreover, the binding between CCL5 mRNA and HNRNPL was also reduced by knockdown of circMGA (Fig. S4C). However, neither HNRNPL nor circMGA was affected by CCL5 overexpression (Fig. S4D, E). To further determine the role of circMGA/HNRNPL complex in regulating the expression of CCL5, we performed HNRNPL knockdown in circMGA-overexpressed bladder cancer cells. CircMGA increased the expression and secretion of CCL5, which was reversed by knockdown of HNRNPL ( Fig. 5D-F), suggesting that the up-regulation of CCL5 by circMGA was dependent on HNRNPL and that circMGA played a vital role in synergizing with HNRNPL to regulate CCL5 expression. Collectively, we demonstrated that circMGA enhanced the expression and secretion of CCL5 through HNRNPL.
CircMGA/HNRNPL complex mediates recruitment of CD8 + T cells Previous studies have shown that CD8 + T cells infiltration required tumor cell-derived CCL5 in solid tumors [7,29]. Hence, we explored whether circMGA/HNRNPL complex could recruit CD8 + T cells into tumors to promote immune attack. To this end, we cocultured bladder cancer cells with CD8 + T cells in a transwell coculturing system, which eliminated direct cell-to-cell interactions, ensuring T cells interacted merely with the tumor-secreted cytokines (Fig. 6A). As demonstrated in the results, overexpression of circMGA could stimulate CD8 + T cells chemotaxis and subsequently enhance their killing effect on tumor cells (Fis. 6B, C). Similar results were also revealed in bladder cancer cells overexpressing HNRNPL (Fig. 6D, E). We then investigated the interplay between circMGA and HNRNPL in regulating the recruitment of CD8 + T cells. Notably, the increase of CD8 + T cells recruitment and antitumor immunity induced by enforced expression of circMGA was attenuated by knockdown of HNRNPL (Fig. 6F, G). In addition, we detected the apoptosis of EJ and CD8 + T cells in vitro using Flow cytometry assay. As the results showed that knockdown of HNRNPL abolished circMGA-mediated apoptosis of bladder cancer cells (Fig. S5A). Meanwhile, there was no significant difference in apoptosis rate of CD8 + T cells between different groups (Fig. S5B).
To confirm whether the effect of circMGA/HNRNPL complex on immune attack was exerted via CCL5, we conducted a series of rescue experiments. It was indicated that knockdown of CCL5 abolished circMGA-mediated recruitment of CD8 + T cells and the tumor inhibition (Fig. 6H, I). Similarly, HNRNPL-mediated promotion of CD8 + T cells infiltration and tumor killing effect was also Fig. 2 CircMGA increases the accumulation of CCL5. A A heatmap of mRNAs in EJ, T24T and UMUC3 cells transfected with vector or circMGA overexpression plasmid. Filtered by log 2 (fold change) ≥ 1 and P < 0.05. Red color represents upregulated mRNAs and blue color represents downregulated mRNAs. B and C Venn diagram show the overlapping of the upregulated (B) or downregulated (C) mRNAs collection in (A). D The mRNA levels in bladder cancer cells with overexpression of circMGA were determined by qRT-PCR. Data were mean ± SD, n = 3. ns means no statistical significance, *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001, Student's t-test. E The expression levels of ATF3 and KLF10 in bladder cancer cells with overexpression of circMGA were detected by western blot. F The expression and secretion of CCL5 in bladder cancer cells with overexpression of circMGA were detected by ELISA. Data were mean ± SD, n = 3. **P < 0.01, ***P < 0.001, Student's t-test. G The Kaplan-Meier curves analyses of OS in bladder cancer patients with low versus high expression of CCL5 from TCGA cohorts. H The expression of CCL5 was detected by qRT-PCR in 73 pairs of bladder cancer and paired adjacent normal bladder tissues. GAPDH was used as internal control. Data were mean ± SD, paired t-test. I OS of bladder cancer patients was analyzed by Kaplan-Meier curves. P value was calculated using a log-rank test. J Analysis of the correlation between the levels of circMGA and CCL5 mRNA in the tumor tissues of the 73 bladder cancer patients. ΔCT values were normalized according to GAPDH. P values were calculated by Pearson correlation analysis.
completely reversed by CCL5 knockdown (Fig. 6J, K). Taken together, the above results suggested that circMGA/HNRNPL complex enabled CD8 + T cells infiltration and exerted immune killing effect on bladder cancer cells through CCL5.

CircMGA/HNRNPL complex suppresses bladder cancer progression in vivo
To determine the significance of circMGA/HNRNPL complex in vivo, subcutaneous bladder cancer model was established in   In addition, knockdown of HNRNPL abolished circMGA-mediated infiltration of CD8 + T cells (Fig. 7D). Moreover, a more intense TUNEL staining in circMGA group compared with vector group was detected, which were attenuated by HNRNPL knockdown (Fig.  7D). Collectively, these data indicated that circMGA/HNRNPL complex suppressed bladder cancer progression by recruiting CD8 + T cells into tumor in vivo.
CircMGA potentiates response to PD-1 checkpoint blockade in bladder cancer Immune checkpoint inhibitors (ICIs), with anti-PD-1/PD-L1 antibodies, have been the primary focus of investigation for the treatment of many cancers, including bladder cancer [30,31].
Cancers infiltrated with CD8 + T cells are associated with a higher likelihood of response to PD-1/PD-L1 blockade [30]. We discovered that the application of anti-PD-1 antibody, in the absence of CD8 + T cells, had no obvious effect on bladder cancer cells viability in vitro (Fig. 7E). However, the administration of anti-PD-1 with the assistance of CD8 + T cells suppressed the growth of tumor cells and circMGA could significantly improve the effect of anti-PD-1 antibody (Fig. 7F, G). To gain further insights into the potential synergistic therapeutic effect of circMGA and anti-PD-1 on bladder cancer in vivo, EJ and T24T cells stably transfected with circMGA or control vector were injected subcutaneously into Hu-HSC-NPG mice, followed by intraperitoneal IgG or anti-PD-1 treatment ( Fig.  7H; Fig. S6F, G). Supporting the results obtained in vitro, as shown in Fig. 7I-K, anti-PD-1 decreased the tumor volume and weight. More importantly, the tumor volume and weight were observed to be strikingly decreased when tumors overexpressing circMGA were treated with anti-PD-1 ( Fig. 7I-K; Fig. S6H-J). Immunohistochemical staining verified that intratumoral human CD8 + T cells were significantly increased by anti-PD-1 antibody, and that its combination with circMGA could significantly increase CD8 + T cells infiltration (Fig. 7L). Strikingly, our findings emphasized that a more obvious improvement of TUNEL staining in the group of circMGA followed with anti-PD-1 treatment, compared with the administration of anti-PD-1 alone (Fig. 7L). In summary, the above results indicated that circMGA/HNRNPL complex hindered tumor progression by augmenting tumor infiltration of CD8 + T cells (Fig.  8), highlighting that therapeutic synergy between circMGA and anti-PD-1 was expected to become a new combination therapy in bladder cancer patients.

DISCUSSION
Bladder cancer is the most prevalent urinary system malignancy worldwide and carries a large societal burden due to required clinical surveillance and multiple therapeutic interventions [1,2]. More recently, immunotherapies targeting the PD-1/PD-L1 are expected to revolutionize the treatment paradigm of bladder cancer [30][31][32]. Nevertheless, how to improve immunotherapy response and what are the roles of circRNAs in immunotherapy are still largely unknown. In this study, we unraveled that circMGA interacted with HNRNPL to promote the stability of CCL5 mRNA, which enhanced recruitment of CD8 + T cells and consequently boosted the efficacy of anti-PD-1 immunotherapy. Hence, we illustrated one of the initial and important mechanisms through which intratumoral CD8 + T-cell infiltration could be increased, and our findings implied that circMGA/HNRNPL complex was a promising target for combination immunotherapy approaches with PD-1 checkpoint blockade.
Covalently closed circRNAs are produced by precursor mRNA back-splicing of thousands of genes in eukaryotes, which can be mediated by the protein QKI, FUS or NF90 [11,33]. Despite that the mechanisms underlying formation of circMGA remain to be elucidated, circMGA was not simply a by-product of splicing but was functional. It has been proposed that most of the exonderived circRNAs are primarily located in cytoplasm and usually function as miRNA response elements [34]. Intriguingly, we ruled out the function of circMGA as miRNA sponge and verified that circMGA performed its protein binding role in the cytoplasm, which contributed to the stabilizing effect of HNRNPL on CCL5 mRNA. In this study, we demonstrated that circMGA could interact with the RRM3-RRM4 tandem domain of HNRNPL and CCL5 3′-UTR bind with the RRM2 tandem domain of HNRNPL. Notably, we found that circMGA could promote the binding of HNRNPL to the CCL5 3′-UTR, which indicated that circMGA might act as a molecular chaperone and possibly change their spatial conformation, thereby promoting protein function. However, the topological structure of circMGA/HNRNPL complex still needs to be further characterized, which may reveal detailed features of this interaction and unveil whether circMGA plays an important role in the conformational change of HNRNPL. Moreover, circRNAs are more stable than their cognate linear transcripts owing to circular structure [33], which is consistent with our findings of circMGA. Regardless of the innate stability of circRNAs, recent studies have shown that they undergo degradation in normal and stress conditions [35,36]. Herein, HNRNPL was responsible for the stability of circMGA to prevent degradation, expanding the regulators of circRNAs stability. While we speculated that the stabilizing effect of HNRNPL on circMGA might be similar to that on CCL5, circMGA and HNRNPL formed a feedback loop that Fig. 3 CircMGA interacts with HNRNPL protein in bladder cancer. A RNA immunoprecipitation (RIP) assays in bladder cancer cells using AGO2 and IgG antibody. The precipitate was subjected to western blotting with the antibody against AGO2. The AGO2-enriched circMGA relative to the IgG-enriched value was calculated by qRT-PCR. Data were mean ± SD. ns means no statistical significance, Student's t-test. B Identification of proteins that interact with circMGA by silver staining. Biotin-labeled sense or antisense circMGA probes were used for RNAprotein pull-down against T24T and UMUC3 cells lysates. Red arrow indicated the major differential band precipitated. C Analysis pipeline was performed to identify proteins that interact with circMGA: (1) The 45 and 45 differential proteins with molecular masses of 55-70 kDa were selected as the candidates in the precipitated proteins of T24T and UMUC3, respectively; (2) Venn diagram depicted the intersection between the differential precipitated proteins and RNA-Binding Protein DataBase (RBPDB); (3) HNRNPL was selected as it was the only protein with high abundance (no less than 3 peptides). D Mass spectrometry assay showed the identified HNRNPL peptides pulled down by sense circMGA probes. E HNRNPL immunoblot analysis of the biotin-labeled sense and antisense circMGA probes pull-down eluate from lysates of T24T and UMUC3 cells. GAPDH was used as loading control. F RIP assays showed the relationship between HNRNPL and circMGA. Top, The HNRNPLenriched circMGA relative to the IgG-enriched value was calculated by qRT-PCR. Bottom, IP efficiency of HNRNPL antibody shown in western blotting. IgG antibody used as a control. Data were mean ± SD. ****P < 0.0001, Student's t-test. G Dual RNA-FISH and immunofluorescence staining assay showed the co-localization of HNRNPL (green) and circMGA (red) in the cytoplasm of EJ, T24T and UMUC3 cells, with nuclei staining with DAPI (blue). Scale bar,10 μm. H CatRAPID algorithm was used to predict the RNA-protein interaction of circMGA with HNRNPL and schematic diagram revealed the domains of HNRNPL truncations. I Immunoblot analysis with anti-Flag of T24T cells transfected with plasmids encoding Flag-tagged HNRNPL or truncated HNRNPLs. J Relative enrichment of endogenous circMGA in full-length or truncated HNRNPL RIP was measured by qRT-PCR. Data were mean ± SD. ns means no statistical significance, ****P < 0.0001, Student's t-test.
allowed cells to respond timelier to microenvironment by augmenting the function of circMGA/HNRNPL complex. However, whether the function of circMGA might exist in other cell types other than bladder cancer cells needs to be further investigated.
Chemokines function as cues for the coordinated recruitment of immune cells into tumor. CCL5, belonging to the C-C motif chemokine family, is a potent chemoattractant for immune cells [37,38]. There are presently conflicting reports as to how CCL5 affects tumor immunity, with findings suggesting that CCL5 suppresses antitumor immunity through triggering macrophage and regulatory T cells chemoattraction [37,39], whereas other studies emphasizing that antitumor immunity is promoted by recruiting CD8 + T cells [8,9]. Thus, tumor-derived CCL5 may have divergent roles in tumor fate determination depending on the tumor type. In bladder cancer, CCL5 expression is positively correlated with CD8 + T cells infiltration [29]. Consistently, we validated that CCL5 stabilized by circMGA/HNRNPL complex was chemotactic for CD8 + T cells. Besides, CCL5 expression was lower in bladder cancer compared with normal bladder tissue, implying that low expression of CCL5 should be added to the molecular alterations that lead to tumor immune escape and predict resistance to immune checkpoint blockade. Our study provided a novel insight into CD8 + T cells recruitment involved in the posttranscriptional regulation of CCL5.
PD-1/PD-L1 checkpoint blockade is a promising clinical anticancer treatment modality by blocking the binding of PD-L1 on tumor cells to PD-1 on activated T cells to reactivate T-cell mediated antitumor immunity [32]. The U.S. Food and Drug Administration (FDA) granted accelerated approval to PD-1/PD-L1 checkpoint blockade five years ago, for the treatment of patients with MIBC who are not eligible for cisplatin-containing chemotherapy. Unfortunately, indications of agents targeting PD-1/PD-L1 for bladder cancer are recently withdrawn because the phase III clinical trials do not meet the primary endpoints [31,40]. The limited efficacy of anti-PD-1/PD-L1 therapy is mainly due to the lack of infiltrating CD8 + T lymphocytes, leading to the inability of the immune system to specifically recognize and respond to the cancer cells [41]. The data in this study showed that circMGA/HNRNPL complex enhanced the immunotherapy efficacy of anti-PD-1 by increasing the recruitment of CD8 + T cells. Of note, circMGA was synergistic with PD-1 blockade in humanized mouse model which harness the human immune system against human tumors. Therefore, these results highlighted that circMGA was a new therapeutic target for breaking out of the current dilemma of immunotherapy for bladder cancer. Additionally, it is of great significance, in the near future, to explore whether the expression of circMGA could be used as a stratification biomarker to select bladder cancer patients who may respond and benefit from PD-1/PD-L1 checkpoint blockade.
In conclusion, we provide the first line of comprehensive evidence that circMGA/HNRNPL complex, forming a positive feedback loop, is a critical tumor suppressor by recruiting CD8 + T cells and boosting the efficacy of immunotherapy in bladder cancer. The insights obtained from this study further advanced our understanding of the physiological roles of circRNAs in antitumor immunity. Remarkably, our findings suggest promising combination therapeutic strategies that circRNAs could be synergistic with immune checkpoint inhibitors.  Supplementary Table S1. Overall survival (OS) was defined as the time between primary surgery and death of any cause or between surgery and the last observation point. Patients who were alive at the end of followup were censored. Follow-up was primarily done by telephone interviews to the patients at predefined intervals. Patients who had received a neoadjuvant chemotherapy or intravesical therapy with Bacillus Calmette-Guerin (BCG) or mitomycin prior to surgery were excluded. Additional reasons for exclusion were patients who lost to follow-up. Demographic characteristics (age, sex), clinicopathological parameters (clinical stages, pathological grades, lymph node metastasis, and blood vessel invasion) and OS were documented and analyzed for correlation with gene expression.

MATERIALS AND METHODS Human tissue specimens and cell cultures
Human immortalized uroepithelium cell line (SV-HUC-1) and bladder cancer cell line UMUC3 were purchased from American Type Culture Collection (ATCC, USA). The human bladder carcinoma EJ cell line was generously provided by Dr.Wun-Jae Kim (Department of Urology, Chungbuk National University, Chungbuk, South Korea). The human metastatic bladder cancer cell line T24T was obtained from Dr. Dan Theodorescu (Departments of Urology, University of Virginia, Charlottesville, VA) [42]. SV-HUC-1 cells were cultured in F-12 K medium (Gibco, USA) supplemented with 10% FBS (Gibco, Australia origin), 1% penicillin/ streptomycin, and 2 mM L-glutamine (Life Technologies, Carlsbad, CA, USA). EJ, T24T and UMUC3 cells were cultured in Dulbecco's modified Eagle's medium (Thermo Scientific, USA) and supplemented with 10% FBS and 1% penicillin/streptomycin. All cell lines were confirmed 4-6 months before use by using a short tandem repeat method and were tested negative for mycoplasma contamination. All cells were grown in an incubator at 37°C with humidified atmosphere of 5% CO 2 .
RNA preparation, RNase R treatment, and qRT-PCR Total RNA from cells or tissue samples was extracted by RNeasy MinElute Cleanup Kit (Qiagen, Germany) according to the manufacturer's instructions. Nuclear and cytoplasmic RNAs were extracted with Cytoplasmic and Nuclear RNA Purification Kit (Norgen Biotek, Thorold, ON, USA). For RNase R treatment, 1 μg of total RNA was incubated 15 min at 37°C with or without 3 U of RNase R (Epicentre Technologies, Madison, WI). RNA samples were synthesized cDNA with random primer (Takara, Japan). Genomic DNA (gDNA) was extracted using Genomic DNA Isolation Kit (Sangon Biotech, Shanghai, China). The cDNA and gDNA PCR products were observed using 1.5% agarose gel electrophoresis. Real-time PCR was performed using AceQ qPCR SYBR Green Master Mix (Vazyme, Nanjing, China). The circRNA and mRNA levels were normalized by GAPDH. The results were analyzed with the Step OnePlus Real-Time PCR System (Applied Biosystems, USA). The 2 −△△Ct method was used to quantify the transcript levels. All primers sequences were listed in Table S2.
Actinomycin D assay and RNA stability assay for RNA lifetime Bladder cancer cells were planted into six-well plates. Up to 60% confluency after 24 h, cells were treated with 5 μg/ml actinomycin D (Sigma-Aldrich, USA) or DMSO (Sigma-Aldrich, USA) and harvested at indicated time points (0, 4, 8, 12, 16, 20, and 24 h). The turnover rate and half-life of RNA were estimated according to a previously published paper [14]. Fig. 4 HNRNPL enhances the stability of circMGA and CCL5. A The RNA levels in bladder cancer cells with or without overexpression of HNRNPL were determined by qRT-PCR. Data were mean ± SD, n = 3. ***P < 0.001, ****P < 0.0001, Student's t-test. B The RNA levels in bladder cancer cells with or without knockdown of HNRNPL were determined by qRT-PCR. Data were mean ± SD, n = 3. **P < 0.01, ***P < 0.001, ****P < 0.0001, Student's t-test. C The expression and secretion of CCL5 in bladder cancer cells with or without overexpression of HNRNPL were detected by ELISA. Data were mean ± SD, n = 3. ***P < 0.001, ****P < 0.0001, Student's t-test. D The expression and secretion of CCL5 in bladder cancer cells with or without knockdown of HNRNPL were detected by ELISA. Data were mean ± SD, n = 3. **P < 0.01, ***P < 0.001, Student's ttest. E Analysis for RNA levels of circMGA by qRT-PCR in bladder cancer cells with or without overexpression of HNRNPL upon treatment with ActD at the indicated time points (n = 3). F Analysis for mRNA levels of CCL5 by qRT-PCR in bladder cancer cells with or without overexpression of HNRNPL upon treatment with ActD at the indicated time points (n = 3). G Dual-luciferase assay revealing the relative luciferase activities of CCL5 3′-UTR in bladder cancer cells stably transfected with or without HNRNPL. Data were mean ± SD, n = 3. ***P < 0.001, Student's t-test. H RIP assays in bladder cancer cells using HNRNPL and IgG antibody. Top, The HNRNPL-enriched CCL5 relative to the IgG-enriched value was calculated by qRT-PCR. Bottom, IP efficiency of HNRNPL antibody shown in western blotting. IgG antibody used as a control. Data were mean ± SD, n = 3. ***P < 0.001, Student's t-test.

RNA fluorescence in situ hybridization (FISH)
FISH probes for circMGA (Supplementary Table S2) were synthesized by TSINGKE (Wuhan, China). FISH assay was undertaken using Fluorescent In Situ Hybridization kit (RiboBio, Guangzhou, China) according to the manufacturer's protocol. U6 or 18s RNA was taken as control. The images were acquired using a confocal laser scanning microscope (LSM 780, Carl Zeiss).

Clone formation assay, transwell migration and matrigel invasion assays
CircMGA-overexpressed EJ, T24T, and UMUC3 cells and the corresponding control cells were seeded in the 6-well plate with 1000 cells/well. Subsequently, the plates were placed in the incubator with standard culture. Finally, clones were harvested and stained with 0.1% crystal violet (Beyotime, Shanghai, China) when over 50 cells per clone were counted. The migration and matrigel invasion assays were performed to evaluate cell migration and invasion abilities by using transwell chamber as described previously [13]. Briefly, circMGA-overexpressed cells and the corresponding control cells (5 × 10 4 cells per well for migration, and 1 × 10 5 per well for invasion) were seeded to the upper chamber and incubate for 20 h with 1% FBS. For invasion assays, the bottom of the upper chamber was coated with Matrigel (BD Bioscience) and incubate for 24 h.

Flow cytometry assay (FCM) for cell apoptosis and cell cycle analyses
Bladder cancer cells were cultured in 6-well plates at a concentration of 1 × 10 5 cells per well. Cell apoptosis were tested by staining cells with Annexin V-fluorescein isothiocyanate (FITC) and propidium iodide (PI) (BD Biosciences, USA) according to the manufacturer's instructions. Flowjo VX10 software was used for analyzing the results. For cell cycle analyses, cells were harvested and stained with propidium iodide buffer (BD Biosciences, USA) according to the manufacturer's instructions. The ModFit LT software was used for analyzing the results.

RNA sequencing
The gene expression profiles of circMGA-overexpressed EJ, T24T, and UMUC3 cells and the corresponding control cells were determined by UID-RNA-seq (SeqHealth Tech, China). Significant differentially expressed transcripts were screened by log 2 (fold change) ≥ 1 and P < 0.05. The mRNA expression profile data reported in this paper have been deposited in the Gene Expression Omnibus (GEO) database with the number of GSE182310.

Enzyme linked immunosorbent assay (ELISA)
Cell culture supernatants were collected and analyzed for cytokines production using ELISAs according to the manufacturer's instructions. ELISA kit for human CCL5 was from R&D Systems (USA).

Biotin-labeled RNA pull down, silver staining and mass spectrometry analysis
Biotin-labeled oligonucleotide probes targeting junction site of circRNA (named as sense probe, which was complementary to the circularization site and able to pull down the circRNA) and control (named as antisense probe, which had the same sequence as the circRNA) probe (Table S2) were synthesized by TSINGKE (Wuhan, China). RNA pull-down assays were performed as previously described [14]. Silver staining was measured by using the PAGE Gel Silver Staining Kit (Solarbio, Beijing, China) based on the protocol described. Retrieved proteins were detected by mass spectrometry analysis at Novogene Company (Beijing, China). The RNA-Binding Protein DataBase (RBPDB) was used to further identify circRNA binding proteins [43].

Fluorescence immunocytochemical staining
Cells were grown on coverslip and incubated with antibody specific for HNRNPL (sc-32317, Santa Cruz) at 4°C overnight, and incubated with Alexa Fluor 488 goat anti-mouse IgG and DAPI. The images were photographed using a confocal laser scanning microscope (LSM 780, Carl Zeiss).

Dual-luciferase reporter assay
The CCL5 3′-UTR sequence and a fragment of CCL5 mRNA 3′-UTR with deletion of the RNA region (326-402 nucleotides) were cloned into psiCHECK2 plasmid. Cells were seeded in 24-well plates at a density of 5 × 10 3 cells per well and maintained for 24 h. The psiCHECK2 CCL5 3′-UTR reporter vector was transfected with HNRNPL overexpression/knockdown cells or circMGA overexpression cells to determine the 3′-UTR activity of CCL5. After transfection for 48 h, the firefly and Renilla luciferase activities were measured with Dual-Luciferase® Reporter Assay System (Promega, USA) according to the manufacturer's protocol.

Isolation and culture of CD8 + T cells from peripheral blood
The blood was obtained from healthy volunteers, and human peripheral blood mononuclear cells (hPBMCs) were isolated and purified by using the Ficoll kit (Sigma-Aldrich, USA) according to the manufacture's instruction. Fig. 5 CircMGA promotes the stability of CCL5 mRNA through HNRNPL. A Analysis for mRNA levels of CCL5 by qRT-PCR in bladder cancer cells with or without overexpression of circMGA upon treatment with ActD at the indicated time points (n = 3). B Dual-luciferase assay revealed the relative luciferase activities of CCL5 3′-UTR in bladder cancer cells stably transfected with or without circMGA. Data were mean ± SD, n = 3. ***P < 0.001, ****P < 0.0001, Student's t-test. C Relative enrichment of endogenous CCL5 mRNA in RIP analysis, following bladder cancer cells transfected with or without circMGA. Top, The HNRNPL-enriched CCL5 relative to the IgG-enriched value was calculated by qRT-PCR. Bottom, IP efficiency of HNRNPL antibody shown in western blotting. IgG antibody used as a control. Data were mean ± SD, n = 3. **P < 0.01, ***P < 0.001, ****P < 0.0001, Student's t-test. D The CCL5 mRNA levels in bladder cancer cells stably transfected with scramble or knockdown HNRNPL, and those co-transfected with vector or circMGA. Data were mean ± SD, n = 3. *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001, Student's t-test. E The protein levels of CCL5 detected by ELISA in bladder cancer cells stably transfected with empty or knockdown HNRNPL, and those co-transfected with vector or circMGA. Data were mean ± SD, n = 3. *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001, Student's t-test. F Dual-luciferase assay revealing the relative luciferase activities of CCL5 3′-UTR in bladder cancer cells stably transfected with empty vector or knockdown HNRNPL, and those co-transfected with vector or circMGA. Data were mean ± SD, n = 3. **P < 0.01, ***P < 0.001, ****P < 0.0001, Student's t-test.

Chemotaxis assay
Chemotaxis of CD8 + T lymphocyte cells was analyzed in a transwell system (Corning, Tewksbury, MA, USA) using 5-μm polycarbonate membranes [44]. Prior to the assay, bladder cancer cells were added in lower chamber. After incubation at 37°C for 24 h, CD8 + T lymphocytes (5 × 10 5 cells/200 µl) were added to the upper chambers and incubated for 4 h. Following incubation, migrated CD8 + T cells were harvested from the lower chambers and stained with Zombie Aqua™ Fixable Viability Kit (423101, BioLegend, USA) and PerCP-Cy™5.5 Mouse Anti-Human CD8 (560662, BD Biosciences). After staining, 100 µl of Precision Count Beads (424902, Biolegend, USA) were added, and the numbers of migrated CD8 + T cells were determined using flow cytometry (BD Biosciences). Signals of samples were recorded by flow cytometer. More than 5000 cells were acquired for each sample. Recorded results in FCS files were imported into Flowjo software. Samples were gated for singlets by FSC-A/SSC-A discrimination, and live cell population was gated on the basis of zombie violet negative staining, followed by gating on CD8 population. Absolute cell counts were calculated using the following formula: Absolute cell countðcells=μlÞ ¼ cell count precision count beads count precision count beads concentration ðbeads=μlÞ: In addition, the apoptosis of tumor cells and CD8 + T cells was detected by staining cells with Annexin V-fluorescein isothiocyanate (FITC) and propidium iodide (PI) (BD Biosciences, USA) according to the manufacturer's instructions.

CCK8 assay
The proliferation of cells was tested by CCK8 kit (Dojindo, Japan) following the manufacturer's instructions. Briefly, EJ cells transfected with plasmids were seeded in 96-well plates and cultured for different time periods respectively (0, 24, 48, and 72 h). For chemotaxis assay, CD8 + T lymphocytes cells in the upper chambers were removed, then bladder cancer cells in the lower chamber following chemotaxis were continue cultured for different time periods respectively (0, 24, 48, and 72 h). The optical density (OD) at 450 nm was measured using an automatic microplate reader (Synergy4; BioTek, Winooski, VT, USA). The proliferation index was defined as the ratio of the sample's OD value at each time point to that at 0 h.

Animal experiments
In order to better simulate the human immune system, human immune system was reconstituted in NPG (NOD-protein kinase DNA-activated catalytic severe combined immune deficiency Il2rg null ) mice by transplanting human hematopoietic stem cells (Hu-HSC-NPG mice). Hu-HSC-NPG female mice of 16-17 weeks of age weighing 18-22 g were obtained from Vitalstar Biotechnology Co., Ltd (Beijing, China). Cells of the peripheral blood were stained with anti-mouse CD45 (147710, Biolegend) and anti-human CD45 (304012, Biolegend) antibodies to distinguish between mouse and human cells. Mice were considered humanized if human CD45 reached~25% or more among CD45 + (human CD45 + and mouse CD45 + ) lymphocytes in circulating blood of reconstituted Hu-HSC-NPG mice. All reconstituted mice were random assigned into four experimental groups (n = 6 per group). All mice were maintained in a special pathogen-free environment and were used with the approval of the Animal Ethics Committee of Tongji Medical College of Huazhong University of Science and Technology. Mice were subcutaneously injected with EJ or T24T cells (5 × 10 6 ) both the left and right flank. To explore potential synergistic therapeutic effect of circMGA and anti-PD-1 in vivo, all tumors were treated with intraperitoneal injection of anti-PD-1 antibody (Tislelizumab, BeiGene, China) at the dosage of 100 μg per mouse every 4 days for seven times and respective IgG antibody was used as control at similar dosage and frequency. The lengths (major tumor axis) and widths (minor tumor axis) of the tumors were measured with a caliper every week, and the tumor volumes were calculated with the following formula: V = 0.5 × length × (width) 2 . At day 28 or 42, animals were sacrificed under anesthesia, after which tumors were harvested and immediately snap-frozen in liquid nitrogen. All experiments were performed in accordance with protocols approved by Institutional Animal Care and Use Committee Tongji Medical College, Huazhong University of Science and Technology.

Statistics
For statistical data analysis, GraphPad Prism 7.0 software (Graphpad software, San Diego, CA) was used. Descriptive statistics were summarized as frequencies with percentages for categorical values, and medians with interquartile ranges or means ± standard deviation (SD) for continuous variables, depending on the data distribution. Comparison of variables with normal distributions was tested using two-tailed Student's t test. The Wilcoxon rank-sum test was used for nonparametric variables. The Chi-square test was used to evaluate the association of the expression of circMGA with the patient's clinicopathological characteristics. OS distributions of bladder cancer patients with different expression levels of circMGA, mMGA and CCL5 were assessed with the Kaplan-Meier method and compared by the Log-rank test. Correlations were analyzed by Pearson's correlation test. All experiments were carried out at least three times and data from one representative experiment are shown. P values < 0.05 were considered statistically significant. Fig. 6 CircMGA/HNRNPL complex mediates recruitment of CD8 + T cells. A The schematic diagram of the transwell co-culturing system. Preactivated CD8 + T cells were added to the upper chamber, and bladder cancer cells were added to the lower chamber. B Flow cytometry identification of migrated CD8 + T cells count in the lower chamber using Precision Count Beads after 4 h of co-culture. Bladder cancer cells in lower chamber were transfected with vector or circMGA. Data were mean ± SD, n = 3. **P < 0.01, Student's t-test. C The viability of bladder cancer cells overexpressing circMGA in co-culturing system were measured by CCK8. Data were mean ± SD, n = 3. **P < 0.01, ***P < 0.001, Student's t-test. D Flow cytometry identification of migrated CD8 + T cells count in the lower chamber using Precision Count Beads after 4 h of co-culture. Bladder cancer cells in lower chamber were transfected with empty or HNRNPL. Data were mean ± SD, n = 3. ***P < 0.001, ****P < 0.0001, Student's t-test. E The viability of bladder cancer cells overexpressing HNRNPL in co-culturing system were measured by CCK8. Data were mean ± SD, n = 3. ***P < 0.001, ****P < 0.0001, Student's t-test. F Flow cytometry identification of migrated CD8 + T cells count in the lower chamber using Precision Count Beads after 4 h of co-culture. Bladder cancer cells in lower chamber were transfected with vector or circMGA and co-transfected with scramble or shHNRNPL#2. Data were mean ± SD, n = 3. **P < 0.01, ***P < 0.001, ****P < 0.0001, Student's ttest. G The viability of bladder cancer cells transfected with vector or circMGA and co-transfected with scramble or shHNRNPL#2 in coculturing system were measured by CCK8. Data were mean ± SD, n = 3. *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001, Student's t-test. H Flow cytometry identification of migrated CD8 + T cells count in the lower chamber using Precision Count Beads after 4 h of co-culture. Bladder cancer cells in lower chamber were transfected with vector or circMGA and co-transfected with scramble or shCCL5. Data were mean ± SD, n = 3. **P < 0.01, ***P < 0.001, ****P < 0.0001, Student's t-test. I The viability of bladder cancer cells transfected with vector or circMGA and co-transfected with scramble or shCCL5 in co-culturing system were measured by CCK8. Data were mean ± SD, n = 3. *P < 0.05, **P < 0.01, ***P < 0.001, Student's t-test. J Flow cytometry identification of migrated CD8 + T cells count in the lower chamber using Precision Count Beads after 4 h of co-culture. Bladder cancer cells in lower chamber were transfected with empty or HNRNPL and co-transfected with scramble or shCCL5. Data were mean ± SD, n = 3. **P < 0.01, ***P < 0.001, ****P < 0.0001, Student's t-test. K The viability of bladder cancer cells transfected with empty or HNRNPL and co-transfected with scramble or shCCL5 in co-culturing system were measured by CCK8. Data were mean ± SD, n = 3. *P < 0.05, **P < 0.01, ***P < 0.001, Student's t-test. Fig. 7 Biological implications of circMGA in bladder cancer. The volume (A, B) and weight (C) of subcutaneous xenograft tumors of EJ cells (n = 6 mice per group). Data were mean ± SD, **P < 0.01, ****P < 0.0001, Student's t-test. D IHC of CD8 and TUNEL in the subcutaneous tumors. Scale bar, 50 µm. E The effect of anti-PD-1 antibody on viability of EJ cells that were not co-cultured with CD8 + T cells in vitro. Data were mean ± SD, ns means no statistical significance, Student's t-test. F The schematic diagram of the transwell co-culturing system. Preactivated CD8 + T cells were added to the upper chamber, and EJ cells were added to the lower chamber. After 4 h of co-culture, the upper chamber was removed from the transwell co-culturing system, and anti-PD-1 was added to the medium of lower chamber. G The viability of EJ cells transfected with vector or circMGA upon treatment with IgG or anti-PD-1 in the transwell co-culturing system were measured by CCK8. Data were mean ± SD, **P < 0.01, ****P < 0.0001, Student's t-test. H Schema for anti-PD-1 therapy. Hu-HSC-NPG mice injected with EJ cells overexpressing vector or circMGA subcutaneously (5 × 10 6 cells on each side of mouse). After 2 weeks, all tumors were treated with intraperitoneal injection of anti-PD-1 antibody at the dosage of 100 μg per mouse every 4 days for seven times, and respective IgG antibody was used as control at similar dosage and frequency. The volume (I, J) and weight (K) of subcutaneous xenograft tumors (n = 6 mice per group). Data were mean ± SD, **P < 0.01, ***P < 0.001, Student's t-test. L IHC of CD8 and TUNEL in the subcutaneous tumors. Scale bar, 50 µm. Fig. 8 Schematic model for the mechanisms of circMGA/HNRNPL complex in hindering bladder cancer progression. As a cytoplasm circRNA, circMGA interacted with HNRNPL to promote its binding to CCL5 mRNA, thereby increasing the stability of CCL5 mRNA and augmenting the recruitment of CD8 + T cells and the efficacy of immunotherapy. In turn, HNRNPL could also increase the stability of circMGA, forming a positive feedback loop that enhanced the function of circMGA/HNRNPL complex.